DICOOGLE: No-SQL for supporting Big Data environments

• Autor(a) principal: Alves, André Filipe Pereira
• Data de Publicação: 2016
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/17218
• topic: Engenharia de computadores e telemática; Bioinformática; Sistemas de informação médica; Diagnóstico por imagem; Armazenamento de dados - Imagens; Recuperação da informação; Bases de dados relacionais

The last few years have been characterized by a proliferation of different types of medical imaging modalities in healthcare institutions. As a result, the services are migrating to infrastructures in the Cloud. Thus, in addition to a scenario where tremendous amounts of data are produced, we walked to a reality where processes are increasingly distributed. Consequently, this reality has created new technological challenges regarding storage, management and handling of this data, in order to guarantee high availability and performance of the information systems, dealing with the images. An Open Source Picture Archive and Communication System (PACS) has been developed by the bioinformatics research group at the University of Aveiro labeled Dicoogle. This system replaced the traditional relational database engine for an agile mechanism, which indexes and retrieves data. Thus it is possible to extract, index and store all the image’s metadata, including any private information, without re-engineering or reconfiguration process. Among other use cases, this system has already indexed more than 22 million images in 3 hospitals from the region of Aveiro. Currently, Dicoogle provides a solution based on the Apache Lucene library. However, it has performance issues in environments where we need to handle and search over large amounts of data, more particularly in data analytics scenarios. In the context of this work, different technologies capable of supporting a database of an image repository were studied. In sequence, four solutions were fully implemented based on relational databases, NoSQL and two distinct text engines. A test platform was also developed to evaluate the performance and scalability of these solutions, which allowed a comparative analysis of them. In the end, it is proposed a hybrid architecture of medical image database, which was implemented and validated. This proposal has demonstrated significant gains in terms of query, index time and in scenarios where it is required a wide data analyze.